Song Recommendation with SESRec on CPU

Introduction

This project adapts the SESRec (Search-Enhanced Sequential Recommendation) model originally designed for e-commerce to the domain of song recommendation. It explores the adaptability of this recommendation system to different application fields and hardware configurations.

SESRec Model

SESRec leverages user search interests to enhance sequential recommendations. It disentangles distinct search and recommendation behaviors to provide more accurate recommendations.

Implementation Details

This section provides an overview of the key components of the project:

• Dataset: We utilized the Spotify Playlist Dataset, which contains user interactions with songs. This dataset was preprocessed to align with the requirements of the SESRec model.

• Data Preprocessing: The dataset was cleaned, unnecessary columns were removed, and the data was converted into a numerical format. The data was split into training, test, and validation sets.

• CPU-Based Training: In contrast to the original implementation that often relies on GPU acceleration, we conducted training on a CPU. This approach demonstrated the model's adaptability to resource-constrained environments.

• Evaluation Metrics: We used standard recommendation metrics, including NDCG, Hit Ratio, and MRR, to assess the performance of the model.

Results

The training process led to improvements in the model's recommendation performance over four epochs. Key results include:

• Increasing NDCG values at different cutoffs (e.g., NDCG@5 and NDCG@10), indicating better ranking quality.
• Rising Hit Ratio values (Hit@1, Hit@5, and Hit@10), showing that the model is effectively finding and recommending relevant songs.
• A growing MRR, suggesting the model's ability to rank relevant songs higher in recommendations.

These results collectively demonstrate the model's effectiveness in delivering high-quality song recommendations.

Conclusion

The project showcases the adaptability of SESRec to a new application domain and hardware configuration. By expanding the model's horizons from e-commerce to music, we unlock its potential for personalized music discovery. This adaptation fosters a synergy between technology and artistry, enhancing digital music experiences.

How to Use

• Requirements:

    python==3.8.13
    torch==1.9.0
    numpy==1.23.2
    pandas==1.4.4
    scikit-learn==1.1.2
    tqdm==4.64.0
    PyYAML==6.0

• Installation:

  • Download data

    Download and unzip data from this link.

  • Preprocess Data:

    The data has to be processed in three steps:

    1. Identifying Search and Recommendation Behaviors
    2. Converting to Numerical Format
    3. Splitting the Data

    Preprocessed data can be found under the here

• Training: Run codes in command line:

  python3 main.py --name SESRec --workspace ./workspace/SESRec --epochs 5 --model SESRec --batch_size 256 --dataset_name spotify

• Evaluation:

  Check training and evaluation process:

  After training, check log files, for example, workspace/SESRec/log/default.log.

Citation

@inproceedings{si2023SESRec,
author = {Si, Zihua and Sun, Zhongxiang and Zhang, Xiao and Xu, Jun and Zang, Xiaoxue and Song, Yang and Gai, Kun and Wen, Ji-Rong},
title = {When Search Meets Recommendation: Learning Disentangled Search Representation for Recommendation},
year = {2023},
isbn = {9781450394086},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3539618.3591786},
doi = {10.1145/3539618.3591786},
booktitle = {Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval},
pages = {1313–1323},
numpages = {11},
keywords = {search, contrastive learning, disentanglement learning, recommendation},
location = {Taipei, Taiwan},
series = {SIGIR '23}
}

Authors

• Sakshi Sanghavi
• Shivam Patel